MCP server

Connect Claude, Cursor, VS Code or any MCP client to OKYC — create sessions, review cases, build workflows and read the docs from your AI assistant.

OKYC runs a hosted Model Context Protocol server. Once connected, your AI assistant can set up an integration end to end (workflow, test session, webhook), look up and review sessions, and answer questions from these docs.


URL	`https://mcp.okyc.stream/mcp`
Transport	Streamable HTTP (stateless)
Sign-in	OAuth 2.1 with PKCE — you sign in with your OKYC dashboard account
Client registration	Client ID Metadata Documents, or dynamic client registration
Scopes	`okyc:read` (look things up), `okyc:write` (create and change)

Connect

The first time you use it, your client opens the OKYC dashboard: sign in (with 2FA if you use it) and approve access. The client then acts as you, with your role, in the organizations you belong to. Tools take optional organization_id and project_id: you only need them when you belong to several organizations or an organization has several projects — list_organizations lists both. A test key already names its project.

Claude Code

claude-code.sh

claude mcp add --transport http okyc https://mcp.okyc.stream/mcp

Claude (web and desktop): Settings → Connectors → Add custom connector, name OKYC, URL https://mcp.okyc.stream/mcp.

Cursor

cursor.mcp.jsonc

{
  "mcpServers": {
    "okyc": { "url": "https://mcp.okyc.stream/mcp" }
  }
}

VS Code

vscode.mcp.jsonc

{
  "servers": {
    "okyc": { "type": "http", "url": "https://mcp.okyc.stream/mcp" }
  }
}

Headless (CI, scripts): instead of signing in, send a test secret key. Live keys are refused — they must never sit in an AI tool's configuration.

headless.mcp.jsonc

{
  "mcpServers": {
    "okyc-sandbox": {
      "url": "https://mcp.okyc.stream/mcp",
      "headers": { "Authorization": "Bearer ${OKYC_TEST_KEY}" }
    }
  }
}

The dashboard's Developers → AI tools (MCP) page shows the same setup, and lists connected tools so you can disconnect them.

Safety

Test mode by default. Every tool takes mode; it is test unless you (or your assistant, because you asked) pass live. Some tools are test-only: running checks directly, trying a draft, creating API keys.
Your role applies. Tools go through the same API as the dashboard: a reviewer cannot publish workflows, a developer cannot decide sessions, audit events are admin-only.
Personal data stays out of AI context. Tool results are sent to your AI provider. In live mode, names, dates of birth, addresses, document numbers and extracted fields are masked unless an admin turns on Share personal data with AI tools (Developers → AI tools). Evidence files are never available.
Not available from AI tools: erasing personal data, creating live API keys, overriding a final decision, and changing settings. Use the dashboard.
Approval before changes. Tools are annotated read-only, write or destructive; MCP clients ask you before running write and destructive tools — keep that on.
Audited. Every call is recorded in the audit log as you, via the MCP client (mcp.tool_called), in addition to what the underlying action records.
Rate limit: 120 tool calls per minute per person or key.

Prompts and resources

Prompt Integrate OKYC into my app — plans and implements an integration for your stack, in test mode.
Prompt Review a session — explains why a session needs review and recommends a decision (without deciding).
Resource okyc://docs/llms.txt — the documentation index (same as https://docs.okyc.stream/llms.txt).

Tools

Tool	Kind	Modes	What it does
`list_organizations`	read-only	—	Organizations you belong to, with your role and each organization’s projects. Pass `organization_id` / `project_id` to other tools when there is more than one.
`get_usage`	read-only	test, live	Checks and sessions used in the last 90 days, by check and provider.
`list_audit_events`	read-only	test, live	Recent audit log (who did what, never values). Admins and owners only.
`list_api_keys`	read-only	test, live	API keys of this mode: name, prefix, last use. Secrets are never returned.
`create_test_api_key`	write	test only	Create a sandbox secret key (sk_test_…) for a backend integration. Returned once. Live keys can only be created in the dashboard.
`create_session`	write	test, live	Create a session for one applicant and get its `url` (open it in a browser or pass it to an SDK). Test mode by default.
`list_sessions`	read-only	test, live	Most recent sessions, newest first. Filter by status (e.g. review, approved), type or workflow.
`get_session`	read-only	test, live	Everything about one session: steps, checks (with `verified_by`), decisions and rules fired, screening hits, manual checks waiting for review, timeline.
`get_review_queue`	read-only	test, live	Sessions waiting for a reviewer, oldest first, with the reasons they need review.
`get_evaluation_preview`	read-only	test, live	What the rules would decide now with current data (after corrections or manual reviews). Records nothing.
`review_manual_check`	write	test, live	For checks verified by a reviewer (provider `manual`): confirm or reject against the uploaded document. Only before a final decision. Find them in get_session → manual_checks.
`decide_session`	destructive	test, live	Approve, decline or request resubmission of a session waiting in `review`. Final decisions cannot be overridden from an AI tool (use the dashboard).
`cancel_session`	destructive	test, live	Cancel an active session; its link stops working.
`list_workflow_templates`	read-only	test, live	Ready-made workflow definitions (e.g. India KYC standard) to start from.
`list_workflows`	read-only	test, live	Workflows of the organization with their current version.
`get_workflow`	read-only	test, live	A workflow with its current definition and version history.
`get_verification_setup`	read-only	test, live	Which steps are verified manually by a reviewer (no registry connected) and which integrations (DigiLocker, phone OTP) are configured for live mode.
`validate_workflow`	read-only	test, live	Report every issue (errors and warnings, per step) in a definition without saving it.
`list_rule_fields`	read-only	test, live	Variables rules can use for a definition (check statuses, scores, name/DOB matches…).
`create_workflow`	write	test, live	Create and publish a new workflow (version 1). Validate it first with validate_workflow.
`save_workflow_draft`	write	test, live	Save an unpublished draft (shared with the dashboard builder). Returns its validation.
`test_workflow_draft`	write	test only	Create a sandbox session running a draft definition, without publishing it. Test mode only.
`publish_workflow`	write	test, live	Publish a new version; new sessions use it (running sessions keep theirs).
`run_check`	write	test only	Run one check without a session, e.g. in.pan.verify with sandbox magic values. Test mode only (live checks may cost money and must come from your backend).
`list_capabilities`	read-only	test, live	Workflow step capabilities per country and the checks they run.
`list_document_types`	read-only	test, live	Documents the documents step can request, with every extractable field fully described.
`list_webhook_endpoints`	read-only	test, live	Webhook endpoints of this mode and the events they receive.
`create_webhook_endpoint`	write	test, live	Register an HTTPS endpoint for events (e.g. session.approved). Returns its signing secret once.
`delete_webhook_endpoint`	destructive	test, live	Stop sending events to an endpoint.
`list_webhook_deliveries`	read-only	test, live	Recent delivery attempts for an endpoint (status, response code, retries).
`search_docs`	read-only	—	Find documentation pages (integration guides, SDKs, API, concepts) for a question.
`get_doc_page`	read-only	—	A documentation page as Markdown, e.g. /mobile/android or /backend/webhooks.

list_organizations

List organizations. Organizations you belong to, with your role and each organization’s projects. Pass organization_id / project_id to other tools when there is more than one.

No arguments.